Prediction of SAMPL3 host–guest binding affinities: evaluating the accuracy of generalized force-fields

We used the second-generation mining minima method (M2) to compute the binding affinities of the novel host-guest complexes in the SAMPL3 blind prediction challenge. The predictions were in poor agreement with experiment, and we conjectured that much of the error might derive from the force field, CHARMm with Vcharge charges. Repeating the calculations with other generalized force-fields led to no significant improvement, and we observed that the predicted affinities were highly sensitive to the choice of force-field. We therefore embarked on a systematic evaluation of a set of generalized force fields, based upon comparisons with PM6-DH2, a fast yet accurate semi-empirical quantum mechanics method. In particular, we compared gas-phase interaction energies and entropies for the host-guest complexes themselves, as well as for smaller chemical fragments derived from the same molecules. The mean deviations of the force field interaction energies from the quantum results were greater than 3 kcal/mol and 9 kcal/mol, for the fragments and host-guest systems respectively. We further evaluated the accuracy of force-fields for computing the vibrational entropies and found the mean errors to be greater than 4 kcal/mol. Given these errors in energy and entropy, it is not surprising in retrospect that the predicted binding affinities deviated from the experiment by several kcal/mol. These results emphasize the need for improvements in generalized force-fields and also highlight the importance of systematic evaluation of force-field parameters prior to evaluating different free-energy methods.     

Computational fragment-based screening using RosettaLigand: the SAMPL3 challenge

SAMPL3 fragment based virtual screening challenge provides a valuable opportunity for researchers to test their programs, methods and screening protocols in a blind testing environment. We participated in SAMPL3 challenge and evaluated our virtual fragment screening protocol, which involves RosettaLigand as the core component by screening a 500 fragments Maybridge library against bovine pancreatic trypsin. Our study reaffirmed that the real test for any virtual screening approach would be in a blind testing environment. The analyses presented in this paper also showed that virtual screening performance can be improved, if a set of known active compounds is available and parameters and methods that yield better enrichment are selected. Our study also highlighted that to achieve accurate orientation and conformation of ligands within a binding site, selecting an appropriate method to calculate partial charges is important. Another finding is that using multiple receptor ensembles in docking does not always yield better enrichment than individual receptors. On the basis of our results and retrospective analyses from SAMPL3 fragment screening challenge we anticipate that chances of success in a fragment screening process could be increased significantly with careful selection of receptor structures, protein flexibility, sufficient conformational sampling within binding pocket and accurate assignment of ligand and protein partial charges.     

Prediction of hydration free energies for the SAMPL4 data set with the AMOEBA polarizable force field

Hydration free energy calculations are often used to validate molecular simulation methodologies and molecular mechanics force fields. We use the free-energy perturbation method together with the AMOEBA polarizable force field and the Poltype parametrization protocol to predict the hydration free energies of 52 molecules as part of the SAMPL4 blind challenge. For comparison, similar calculations are performed using the non-polarizable General Amber force field. Against our expectations, the latter force field gives the better results compared to experiment. One possible explanation is the sensitivity of the AMOEBA results to the conformation used for parametrization.     

Blind prediction of cyclohexane–water distribution coefficients from the SAMPL5 challenge

In the recent SAMPL5 challenge, participants submitted predictions for cyclohexane/water distribution coefficients for a set of 53 small molecules. Distribution coefficients (log D) replace the hydration free energies that were a central part of the past five SAMPL challenges. A wide variety of computational methods were represented by the 76 submissions from 18 participating groups. Here, we analyze submissions by a variety of error metrics and provide details for a number of reference calculations we performed. As in the SAMPL4 challenge, we assessed the ability of participants to evaluate not just their statistical uncertainty, but their model uncertainty—how well they can predict the magnitude of their model or force field error for specific predictions. Unfortunately, this remains an area where prediction and analysis need improvement. In SAMPL4 the top performing submissions achieved a root-mean-squared error (RMSE) around 1.5 kcal/mol. If we anticipate accuracy in log D predictions to be similar to the hydration free energy predictions in SAMPL4, the expected error here would be around 1.54 log units. Only a few submissions had an RMSE below 2.5 log units in their predicted log D values. However, distribution coefficients introduced complexities not present in past SAMPL challenges, including tautomer enumeration, that are likely to be important in predicting biomolecular properties of interest to drug discovery, therefore some decrease in accuracy would be expected. Overall, the SAMPL5 distribution coefficient challenge provided great insight into the importance of modeling a variety of physical effects. We believe these types of measurements will be a promising source of data for future blind challenges, especially in view of the relatively straightforward nature of the experiments and the level of insight provided.     

All-atom/coarse-grained hybrid predictions of distribution coefficients in SAMPL5

We present blind predictions submitted to the SAMPL5 challenge on calculating distribution coefficients. The predictions were based on estimating the solvation free energies in water and cyclohexane of the 53 compounds in the challenge. These free energies were computed using alchemical free energy simulations based on a hybrid all-atom/coarse-grained model. The compounds were treated with the general Amber force field, whereas the solvent molecules were treated with the Elba coarse-grained model. Considering the simplicity of the solvent model and that we approximate the distribution coefficient with the partition coefficient of the neutral species, the predictions are of good accuracy. The correlation coefficient, R is 0.64, 82 % of the predictions have the correct sign and the mean absolute deviation is 1.8 log units. This is on a par with or better than the other simulation-based predictions in the challenge. We present an analysis of the deviations to experiments and compare the predictions to another submission that used all-atom solvent.     

Rapid prediction of solvation free energy. 3. Application to the SAMPL2 challenge

The SAMPL2 hydration free energy blind prediction challenge consisted of a data set of 41 molecules divided into three subsets: explanatory, obscure and investigatory, where experimental hydration free energies were given for the explanatory, withheld for the obscure, and not known for the investigatory molecules. We employed two solvation models for this challenge, a linear interaction energy (LIE) model based on explicit-water molecular dynamics simulations, and the first-shell hydration (FiSH) continuum model previously calibrated to mimic LIE data. On the 23 compounds from the obscure (blind) dataset, the prospectively submitted LIE and FiSH models provided predictions highly correlated with experimental hydration free energy data, with mean-unsigned-errors of 1.69 and 1.71 kcal/mol, respectively. We investigated several parameters that may affect the performance of these models, namely, the solute flexibility for the LIE explicit-solvent model, the solute partial charging method, and the incorporation of the difference in intramolecular energy between gas and solution phases for both models. We extended this analysis to the various chemical classes that can be formed within the SAMPL2 dataset. Our results strengthen previous findings on the excellent accuracy and transferability of the LIE explicit-solvent approach to predict transfer free energies across a wide spectrum of functional classes. Further, the current results on the SAMPL2 test dataset provide additional support for the FiSH continuum model as a fast yet accurate alternative to the LIE explicit-solvent model. Overall, both the LIE explicit-solvent model and the FiSH continuum solvation model show considerable improvement on the SAMPL2 data set over our previous continuum electrostatics-dispersion solvation model used in the SAMPL1 blind challenge.     

Quantum mechanical polarizable force field (QMPFF3): refinement and validation of the dispersion interaction for aromatic carbon

The authors have recently introduced a general, polarizable force field QMPFF fitted solely to high-level quantum mechanical data for simulations of biomolecular systems. Here the authors demonstrate using an advanced version QMPFF3 how the problem of insufficient accuracy of the MP2-based training set for the aromatic carbon atom type can be effectively solved by a simple model correction using state-of-the-art CCSD(T) data. The approach demonstrates excellent transferability, which is confirmed for three phases of matter by accurate calculations of the second virial coefficient for benzene vapor and various properties of liquid benzene and polyaromatic hydrocarbon crystals.     

Predicting cyclohexane/water distribution coefficients for the SAMPL5 challenge using MOSCED and the SMD solvation model

We present blind predictions using the solubility parameter based method MOSCED submitted for the SAMPL5 challenge on calculating cyclohexane/water distribution coefficients at 298 K. Reference data to parameterize MOSCED was generated with knowledge only of chemical structure by performing solvation free energy calculations using electronic structure calculations in the SMD continuum solvent. To maintain simplicity and use only a single method, we approximate the distribution coefficient with the partition coefficient of the neutral species. Over the final SAMPL5 set of 53 compounds, we achieved an average unsigned error of \(2.2\pm 0.2\) log units (ranking 15 out of 62 entries), the correlation coefficient (R) was \(0.6\pm 0.1\) (ranking 35), and \(72\pm 6\,\%\) of the predictions had the correct sign (ranking 30). While used here to predict cyclohexane/water distribution coefficients at 298 K, MOSCED is broadly applicable, allowing one to predict temperature dependent infinite dilution activity coefficients in any solvent for which parameters exist, and provides a means by which an excess Gibbs free energy model may be parameterized to predict composition dependent phase-equilibrium.     

Testing the semi-explicit assembly model of aqueous solvation in the SAMPL4 challenge

Here, we test a method, called semi-explicit assembly (SEA), that computes the solvation free energies of molecules in water in the SAMPL4 blind test challenge. SEA was developed with the intention of being as accurate as explicit-solvent models, but much faster to compute. It is accurate because it uses pre-simulations of simple spheres in explicit solvent to obtain structural and thermodynamic quantities, and it is fast because it parses solute free energies into regionally additive quantities. SAMPL4 provided us the opportunity to make new tests of SEA. Our tests here lead us to the following conclusions: (1) The newest version, called Field-SEA, which gives improved predictions for highly charged ions, is shown here to perform as well as the earlier versions (dipolar and quadrupolar SEA) on this broad blind SAMPL4 test set. (2) We find that both the past and present SEA models give solvation free energies that are as accurate as TIP3P. (3) Using a new approach for force field parameter optimization, we developed improved hydroxyl parameters that ensure consistency with neat-solvent dielectric constants, and found that they led to improved solvation free energies for hydroxyl-containing compounds in SAMPL4. We also learned that these hydroxyl parameters are not just fixing solvent exposed oxygens in a general sense, and therefore do not improve predictions for carbonyl or carboxylic-acid groups. Other such functional groups will need their own independent optimizations for potential improvements. Overall, these tests in SAMPL4 indicate that SEA is an accurate, general and fast new approach to computing solvation free energies.     

Calculating distribution coefficients based on multi-scale free energy simulations: an evaluation of MM and QM/MM explicit solvent simulations of water-cyclohexane transfer in the SAMPL5 challenge

One of the central aspects of biomolecular recognition is the hydrophobic effect, which is experimentally evaluated by measuring the distribution coefficients of compounds between polar and apolar phases. We use our predictions of the distribution coefficients between water and cyclohexane from the SAMPL5 challenge to estimate the hydrophobicity of different explicit solvent simulation techniques. Based on molecular dynamics trajectories with the CHARMM General Force Field, we compare pure molecular mechanics (MM) with quantum-mechanical (QM) calculations based on QM/MM schemes that treat the solvent at the MM level. We perform QM/MM with both density functional theory (BLYP) and semi-empirical methods (OM1, OM2, OM3, PM3). The calculations also serve to test the sensitivity of partition coefficients to solute polarizability as well as the interplay of the quantum-mechanical region with the fixed-charge molecular mechanics environment. Our results indicate that QM/MM with both BLYP and OM2 outperforms pure MM. However, this observation is limited to a subset of cases where convergence of the free energy can be achieved.     

Lessons learned from comparing molecular dynamics engines on the SAMPL5 dataset

We describe our efforts to prepare common starting structures and models for the SAMPL5 blind prediction challenge. We generated the starting input files and single configuration potential energies for the host-guest in the SAMPL5 blind prediction challenge for the GROMACS, AMBER, LAMMPS, DESMOND and CHARMM molecular simulation programs. All conversions were fully automated from the originally prepared AMBER input files using a combination of the ParmEd and InterMol conversion programs. We find that the energy calculations for all molecular dynamics engines for this molecular set agree to better than 0.1 % relative absolute energy for all energy components, and in most cases an order of magnitude better, when reasonable choices are made for different cutoff parameters. However, there are some surprising sources of statistically significant differences. Most importantly, different choices of Coulomb’s constant between programs are one of the largest sources of discrepancies in energies. We discuss the measures required to get good agreement in the energies for equivalent starting configurations between the simulation programs, and the energy differences that occur when simulations are run with program-specific default simulation parameter values. Finally, we discuss what was required to automate this conversion and comparison.     

Extensive all-atom Monte Carlo sampling and QM/MM corrections in the SAMPL4 hydration free energy challenge

We present our predictions for the SAMPL4 hydration free energy challenge. Extensive all-atom Monte Carlo simulations were employed to sample the compounds in explicit solvent. While the focus of our study was to demonstrate well-converged and reproducible free energies, we attempted to address the deficiencies in the general Amber force field force field with a simple QM/MM correction. We show that by using multiple independent simulations, including different starting configurations, and enhanced sampling with parallel tempering, we can obtain well converged hydration free energies. Additional analysis using dihedral angle distributions, torsion-root mean square deviation plots and thermodynamic cycles support this assertion. We obtain a mean absolute deviation of 1.7 kcal mol^?1 and a Kendall’s τ of 0.65 compared with experiment.     

Free-energy perturbation and quantum mechanical study of SAMPL4 octa-acid host–guest binding energies

We have estimated free energies for the binding of nine cyclic carboxylate guest molecules to the octa-acid host in the SAMPL4 blind-test challenge with four different approaches. First, we used standard free-energy perturbation calculations of relative binding affinities, performed at the molecular-mechanics (MM) level with TIP3P waters, the GAFF force field, and two different sets of charges for the host and the guest, obtained either with the restrained electrostatic potential or AM1-BCC methods. Both charge sets give good and nearly identical results, with a mean absolute deviation (MAD) of 4 kJ/mol and a correlation coefficient (R ²) of 0.8 compared to experimental results. Second, we tried to improve these predictions with 28,800 density-functional theory (DFT) calculations for selected snapshots and the non-Boltzmann Bennett acceptance-ratio method, but this led to much worse results, probably because of a too large difference between the MM and DFT potential-energy functions. Third, we tried to calculate absolute affinities using minimised DFT structures. This gave intermediate-quality results with MADs of 5–9 kJ/mol and R ² = 0.6–0.8, depending on how the structures were obtained. Finally, we tried to improve these results using local coupled-cluster calculations with single and double excitations, and non-iterative perturbative treatment of triple excitations (LCCSD(T0)), employing the polarisable multipole interactions with supermolecular pairs approach. Unfortunately, this only degraded the predictions, probably because of a mismatch between the solvation energies obtained at the DFT and LCCSD(T0) levels.     

Measuring experimental cyclohexane-water distribution coefficients for the SAMPL5 challenge

Small molecule distribution coefficients between immiscible nonaqueuous and aqueous phases—such as cyclohexane and water—measure the degree to which small molecules prefer one phase over another at a given pH. As distribution coefficients capture both thermodynamic effects (the free energy of transfer between phases) and chemical effects (protonation state and tautomer effects in aqueous solution), they provide an exacting test of the thermodynamic and chemical accuracy of physical models without the long correlation times inherent to the prediction of more complex properties of relevance to drug discovery, such as protein-ligand binding affinities. For the SAMPL5 challenge, we carried out a blind prediction exercise in which participants were tasked with the prediction of distribution coefficients to assess its potential as a new route for the evaluation and systematic improvement of predictive physical models. These measurements are typically performed for octanol-water, but we opted to utilize cyclohexane for the nonpolar phase. Cyclohexane was suggested to avoid issues with the high water content and persistent heterogeneous structure of water-saturated octanol phases, since it has greatly reduced water content and a homogeneous liquid structure. Using a modified shake-flask LC-MS/MS protocol, we collected cyclohexane/water distribution coefficients for a set of 53 druglike compounds at pH 7.4. These measurements were used as the basis for the SAMPL5 Distribution Coefficient Challenge, where 18 research groups predicted these measurements before the experimental values reported here were released. In this work, we describe the experimental protocol we utilized for measurement of cyclohexane-water distribution coefficients, report the measured data, propose a new bootstrap-based data analysis procedure to incorporate multiple sources of experimental error, and provide insights to help guide future iterations of this valuable exercise in predictive modeling.     

On the fly estimation of host–guest binding free energies using the movable type method: participation in the SAMPL5 blind challenge

We review our performance in the SAMPL5 challenge for predicting host–guest binding affinities using the movable type (MT) method. The challenge included three hosts, acyclic Cucurbit[2]uril and two octa-acids with and without methylation at the entrance to their binding cavities. Each host was associated with 6–10 guest molecules. The MT method extrapolates local energy landscapes around particular molecular states and estimates the free energy by Monte Carlo integration over these landscapes. Two blind submissions pairing MT with variants of the KECSA potential function yielded mean unsigned errors of 1.26 and 1.53 kcal/mol for the non-methylated octa-acid, 2.83 and 3.06 kcal/mol for the methylated octa-acid, and 2.77 and 3.36 kcal/mol for Cucurbit[2]uril host. While our results are in reasonable agreement with experiment, we focused on particular cases in which our estimates gave incorrect results, particularly with regard to association between the octa-acids and an adamantane derivative. Working on the hypothesis that differential solvation effects play a role in effecting computed binding affinities for the parent octa-acid and the methylated octa-acid and that the ligands bind inside the pockets (rather than on the surface) we devised a new solvent accessible surface area term to better quantify solvation energy contributions in MT based studies. To further explore this issue a, molecular dynamics potential of mean force (PMF) study indicates that, as found by our docking calculations, the stable binding mode for this ligand is inside (rather than surface bound) the octa-acid cavity whether the entrance is methylated or not. The PMF studies also obtained the correct order for the methylation-induced change in binding affinities and associated the difference, to a large extent to differential solvation effects. Overall, the SAMPL5 challenge yielded in improvements our solvation modeling and also demonstrated the need for thorough validation of input data integrity prior to any computational analysis.     

Anisotropic nonadditive ab initio force field for noncovalent interactions of H2

A quantum mechanical polarizable force field (QMPFF) has been applied to the noncovalent interactions of molecular hydrogen as well as closed-shell monoatomic species (CSMS): rare gases, alkali cations, and halide anions. The importance of all the main energy components is demonstrated: electrostatics (including penetration effect), exchange repulsion, dispersion, and induction. As the MP2 level of quantum mechanics, which is used to parametrize QMPFF, significantly underestimates the H2-H2 dimer binding energy, the force field was refined using state-of-the-art CCSD(T) data. The approach demonstrates excellent transferability, which is confirmed by accurate reproduction of mixed H2-CSMS dimers and the second virial coefficient of hydrogen vapor.     

SAMPL2 and continuum modeling

An account is given of our contributions to the SAMPL2 challenge for vacuum-water transfer energies. These contributions include different charge sets and radii used with Poisson–Boltzmann continuum theory applied to a single low-energy conformation. A rationale for this approach is given, including a summary of what we have learnt over previous SAMPL events. The results strongly suggest the need for new and repeated experimental measurements, both to clarify what appears to be experimental discrepancies in older measurements and to advance the field in a statistically sound manner.     

Virtual screening with AutoDock Vina and the common pharmacophore engine of a low diversity library of fragments and hits against the three allosteric sites of HIV integrase: participation in the SAMPL4 protein–ligand binding challenge

To rigorously assess the tools and protocols that can be used to understand and predict macromolecular recognition, and to gain more structural insight into three newly discovered allosteric binding sites on a critical drug target involved in the treatment of HIV infections, the Olson and Levy labs collaborated on the SAMPL4 challenge. This computational blind challenge involved predicting protein–ligand binding against the three allosteric sites of HIV integrase (IN), a viral enzyme for which two drugs (that target the active site) have been approved by the FDA. Positive control cross-docking experiments were utilized to select 13 receptor models out of an initial ensemble of 41 different crystal structures of HIV IN. These 13 models of the targets were selected using our new “Rank Difference Ratio” metric. The first stage of SAMPL4 involved using virtual screens to identify 62 active, allosteric IN inhibitors out of a set of 321 compounds. The second stage involved predicting the binding site(s) and crystallographic binding mode(s) for 57 of these inhibitors. Our team submitted four entries for the first stage that utilized: (1) AutoDock Vina (AD Vina) plus visual inspection; (2) a new common pharmacophore engine; (3) BEDAM replica exchange free energy simulations, and a Consensus approach that combined the predictions of all three strategies. Even with the SAMPL4’s very challenging compound library that displayed a significantly lower amount of structural diversity than most libraries that are conventionally employed in prospective virtual screens, these approaches produced hit rates of 24, 25, 34, and 27 %, respectively, on a set with 19 % declared binders. Our only entry for the second stage challenge was based on the results of AD Vina plus visual inspection, and it ranked third place overall according to several different metrics provided by the SAMPL4 organizers. The successful results displayed by these approaches highlight the utility of the computational structure-based drug discovery tools and strategies that are being developed to advance the goals of the newly created, multi-institution, NIH-funded center called the “HIV Interaction and Viral Evolution Center”.     

Assessment of performance of the general purpose polarizable force field QMPFF3 in condensed phase

The recently introduced force field (FF) QMPFF3 is thoroughly validated in gas, liquid, and solid phases. For the first time, it is demonstrated that a physically well-grounded general purpose FF fitted exclusively to a comprehensive set of high level vacuum quantum mechanical data applied as it is to simulation of condensed phase provides high transferability for a wide range of chemical compounds. QMPFF3 demonstrates accuracy comparable with that of the FFs explicitly fitted to condensed phase data, but due to high transferability it is expected to be successful in simulating large molecular complexes.